Modelling Safety Performance Functions and Assessing Highway Safety Manual (HSM) Prediction Models for Four-Lane Undivided Arterials – A Case Study of Kathmandu District

Abstract

Rapid urbanization in Kathmandu district has contributed to a significant rise in traffic volumes, which has contributed to an increase in crash rates, especially on urban arterial roads. This study evaluates the applicability of Highway Safety Manual (HSM) predictive models for local conditions through establishing Safety Performance Functions (SPFs) and calibrating them with crash data of fiscal years 2077/78 to 2079/80. A total of seven four-lane undivided arterial segments were chosen, and crash prediction models for multiple-vehicle and single-vehicle collisions have been generated using negative binomial regression model. Statistical parameters, including Akaike Information criteria (AIC), Mean Absolute Deviation (MAD), and Mean Prediction Bias (MPB), were used to assess the model's performance. Among all the crash prediction models developed, Model III had the lowest AIC value and better overall performance for both multiple vehicle and single vehicle collisions. Crash Modification Factors (CMFs) were used to adjust for deviations from the baseline conditions, and a local calibration factor was computed. The calibration factor of 0.53 was derived using observed and predicted crash frequencies. Afterwards, the 2080/81 crash data was used to evaluate the calibrated models. The obtained R² value of 0.6513 indicates a reasonable degree of model accuracy. The study concludes that jurisdiction-specific models with calibrated parameters significantly improve prediction accuracy and can be instrumental in guiding road safety initiatives in Kathmandu and other similar urban contexts.